This paper investigates an urban Emergency Evacuation Network Design (EEND) problem on a large scale when geographical risk in different areas varies. The decisions to make are (i) determining active shelters, (ii) selecting evacuation routes, and (iii) managing the supply of relief commodities from distribution centers to shelters. A region prone to floods and hurricanes is divided into zones, each with a specific vulnerability risk. For each zone, a risk measure is calculated by combining the risk factors –transporting people and relief commodities and the placement of temporary shelters. The objective is to minimize the maximum risk across the network, ensuring a balanced distribution of risk. A combinatorial scenario planning approach is developed to manage the uncertainty in disaster severity and the evacuee numbers. To incorporate varied geographical risks, a smart hybrid optimization approach as a new solution technique is developed, tuned, and validated to solve the EEND problem. The proposed approach uses directed local search structures designed for the EEND problem and an AI-based self-parameter tuning module, enhancing performance. To extract insights, Rennes, France, is considered a case study. The results indicate a reduction in casualties using a min–max formulation compared to traditional sum-risk objectives. Further, a detailed evacuation plan that increases the number of city regions enhances EEND performance. Practical insights suggest minimizing the number of shelters to the essential capacity needed to host all evacuees, as additional shelters may lead to increased evacuation and supply routes, potentially in areas with higher risk.